Load Balancing in Large Distributed Networks
11 October 2016
UAntwerp, Campus Groenenborger, T.129 - Groenenborgerlaan 171 - 2020 Antwerpen (route: UAntwerpen, Campus Groenenborger
Benny Van Houdt
PhD defence Wouter Minnebo - Faculty of Science, Department of Mathematics and Computer Science
We study distributed strategies to load balance jobs over a fixed set of servers in a network, in order to increase the responsiveness of the system.
In our setup, servers exchange information with each other to negotiate possible job transfers.
The amount of communication allowed between servers plays an important role in the effectiveness of a strategy as exchanging more information makes it easier to redistribute the jobs more evenly.
In this thesis we introduce various strategies that adhere to a predefined communication limit, so we can compare strategies under the condition that their overall communication rate is the same.
Similarly, we also explore the case where the job migration rate is limited.
The methodology used throughout the thesis exists in defining mean field models and proving that they correspond to the proper limit process as the system becomes large.
We mainly focus on strategies for which these models allow us to derive the queue length distributions and mean response time of a job in closed form.